Back to search

FRIMEDBIO-Fri med.,helse,biol

Machine Intelligence in HEADaches - Artificial Intelligence and Machine Learning for Diagnostics and Treatment of Primary Headaches

Alternative title: Kunstig Intelligens ved Hodepine - Maskinlæring som grunnlag for modeller for diagnostikk og behandling ved primære hodepinetilstander

Awarded: NOK 12.5 mill.

Headache disorders such as migraine, tension-type headache and cluster headache are some of the most burdensome diseases worldwide. Yet, they are often neglected, and patients suffer in pain for a long time before they get a correct diagnosis and effective treatment. The headache disorders have complex underlying biological mechanisms. This means that symptoms vary considerably from person to person, and several different types of treatment can be effective. At present, only a fraction of a person’s symptoms are used for diagnostics, and no simple marker indicates what treatment will be effective at the individual level. This makes it very challenging for the clinician to accurately predict optimal choice of therapy, and repeated trial-and-error is often required to achieve acceptable treatment results. The goal of the MI-HEAD project is to utilise machine learning to better understand and improve treatment of patients with headaches. Machine learning is a branch of artificial intelligence, where computers are trained to recognize complex patterns in large amounts of data, and make intelligent decisions based on these. Such techniques are especially useful for medical problems where the biological mechanisms are complex to unravel the subtle patterns that explain why some patients respond to certain treatments. Nevertheless, the use of machine learning in headache medicine is largely unexplored. MI-HEAD will combine Norwegian health register data with detailed data from patient’s smartphone headache diaries to create a highly detailed dataset suited for machine learning. By analysing the large dataset with state-of-the art machine learning models we aim to improve our understanding of why some patients respond to certain types of headache treatments, and others do not. Moreover, we hope to construct tools that may help the clinician with accurate diagnostics, and effectively predict disease progression and treatment responsiveness.

Primary headaches are some of the most burdensome diseases worldwide, yet often neglected and patients suffer with pain for a long time before arriving at correct diagnoses and effective treatments. Primary headaches are characterised by remarkably heterogenous phenotypical, physiological and biological patterns. Yet, at present, only a fraction of these are used for clinical decision making and models of inference of underlying neurobiology and pathophysiology, thus yielding low accuracy. Artificial intelligence (AI) and machine learning (ML) has opened up for increased accuracy of diagnostics and prediction of clinical outcomes in several clinical disciplines, and is especially useful for medical problems where the underlying biological associations are irreducibly distributed across a large number of biophenotypical variables. Nevertheless, the use of AI and ML in headache medicine is largely unexplored. The overall aim of this project is to utilise AI and ML to understand and improve diagnostics, prediction of disease progression and prediction of treatment effect in primary headaches. To address the overall goal and objectives we have defined four work packages (WP), each corresponding to a secondary objective. In WP1 we will use readily available Norwegian health register data to construct ML models to predict diagnosis, disease progression, and treatment effect in primary headache and draw inferences on underlying neurobiology explaining these predictions. In WP2 we will implement a novel mobile health tool to prospectively capture large amounts of data that will complement the data collected WP1. In WP3 we will combine data from WP1 and WP2 to improve the ML models and result in a ML decision support tool that may accurately aid in clinical decision-making in new unseen patients. In WP4 we will conduct a randomised controlled trial to evaluate if the developed ML tool is superior to standard of care, thus saving patient suffering and socioeconomic expenses.

Funding scheme:

FRIMEDBIO-Fri med.,helse,biol